high-signal-chart-workflow

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.90). The SKILL.md explicitly instructs the agent to web-search and download authoritative public datasets via curl (Phase 2: "Web-search for authoritative public datasets... Download via curl" e.g., CDC, EIA) and to POST the rendered chart to the Truesight evaluator and read back its judge outputs (Phase 5), so the agent fetches and interprets open/public third-party content that can drive follow-up actions and code changes.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 1.00). The skill explicitly POSTs the rendered chart to the Truesight evaluator at https://api.truesight.goodeyelabs.com/api/eval/tufte-test during runtime and then uses the evaluator's returned "reasoning" as fix-it instructions to modify chart.py, so the fetched content directly controls the agent's instructions and is a required runtime dependency.